On 2019/10/27 1:15, David Marchand wrote:
On Fri, Oct 25, 2019 at 9:55 PM Yasufumi Ogawa <[email protected]> wrote:
The title does not reflect the observed issue.
I would like to consider to revise it.
I understand that secondary processeses can't be started from a docker
container.
I've confirmed that secondary process can be started as a container app
with SPP, and DPDK v18.08 and v19.08. SPP is a multi-process app
supporting container usecases.
http://git.dpdk.org/apps/spp/
Sorry, I don't understand.
Do you mean the secondary processes can be run from containers without
this patch?
Or once this patch is applied?
Secondary processes can be run from a container. The problem is we
cannot run two or more secondary app containers. It is because each of
secondary app container tries to create its metadata file and the name
is decided with PID, and PID is 1 in all of containers. It means that
all of secondaries try to have the same name of metadata file, and
failed to create the same file. The first container app can create
metadata file with PID 1, but failed to create second one with the same
PID 1.
This patch is to change creating metadata file from using PID 1 to
hostname which is unique in each of secondary containers.
To summarize, we can run just one secondary app container without this
patch, but cannot two or more secondary app containers.
The patch title should reflect this.
On Wed, Jul 24, 2019 at 10:20 AM <[email protected]> wrote:
From: Yasufumi Ogawa <[email protected]>
In secondary_msl_create_walk(), it creates a file for fbarrays with its
PID for reserving unique name among secondary processes. However, it
does not work if secondary is run as app container because each of
containerized secondary has PID 1. To reserve unique name, use hostname
instead of PID because hostname is assigned as a short form of 64
digits full container ID in docker.
Cc: [email protected]
I don't think we want to backport this behavior change.
This issue was included from DPDK v18.08, and some users of SPP are
still using stable 18.11. So, I would appreciate if you agree to backport.
This can be discussed later.
Ok to keep stable in CC: then.
Thanks.
Signed-off-by: Yasufumi Ogawa <[email protected]>
---
lib/librte_eal/linux/eal/eal_memalloc.c | 28 +++++++++++++++++++++++--
1 file changed, 26 insertions(+), 2 deletions(-)
diff --git a/lib/librte_eal/linux/eal/eal_memalloc.c
b/lib/librte_eal/linux/eal/eal_memalloc.c
index 1f6a7c18f..356b304a8 100644
--- a/lib/librte_eal/linux/eal/eal_memalloc.c
+++ b/lib/librte_eal/linux/eal/eal_memalloc.c
@@ -1366,6 +1366,7 @@ secondary_msl_create_walk(const struct rte_memseg_list
*msl,
struct rte_memseg_list *primary_msl, *local_msl;
char name[PATH_MAX];
int msl_idx, ret;
+ char proc_id[33] = { 0 }; /* 32bytes is enough if using hostname */
This variable only makes sense in the if (getpid() == 1) branch,
please move it there, and see below comment about using gethostname().
Sure. It works correctly in secondary app container and I should replace it.
Great, can you send a new version of this patch?
Yes, I will fix it soon.
if (msl->external)
return 0;
@@ -1375,8 +1376,31 @@ secondary_msl_create_walk(const struct rte_memseg_list
*msl,
local_msl = &local_memsegs[msl_idx];
/* create distinct fbarrays for each secondary */
- snprintf(name, RTE_FBARRAY_NAME_LEN, "%s_%i",
- primary_msl->memseg_arr.name, getpid());
+ /* If run secondary in a container, the name of fbarray file should
+ * not be decided with pid because getpid() always returns 1.
+ * In docker, hostname is assigned as a short form of full container
+ * ID. So use hostname as unique ID among containers instead.
I understand this is how it works for docker.
Is this the same in other container environments?
Umm... I've tested on other than docker because I don't have test
environment. I am not sure which of container runtimes should be
supported actually. I think it is enough as the first step to fix this
issue of docker. Moreover, the essential problem is that getpid()
returns 1 in docker or not.
I am also not sure which of environments other than docker should be
supported if necessary. What do you think?
No problem if you don't know.
Thanks.
Yasufumi
--
David Marchand
